Multivariate Echocardiographic Phenotyping of Hypertensive Heart Failure Using Unsupervised Machine Learning: A Pilot Study

The study examined the feasibility of using unsupervised machine learning to analyze echocardiographic data in patients with hypertensive heart failure. A pilot study included 102 patients with 11 echocardiographic variables analyzed using K-means clustering. The analysis identified three distinct phenotypes: the first cluster (50 patients) with preserved ejection fraction and concentric remodeling, the second cluster (37 patients) with dilated ventricle and reduced function, and the third cluster (15 patients) with concentric hypertrophy and intermediate ejection fraction. All echocardiographic variables differed significantly between clusters (p < 0.001). The first cluster showed strong concordance with traditional classification (96%), while the other two clusters demonstrated partial discordance with ejection fraction-based classification. The results suggest that a data-driven approach provides more precise characterization of heart failure phenotypes.